SmartSort-CAM

SmartSort-CAM Documentation

Welcome to the official documentation site for SmartSort-CAM — an end-to-end intelligent computer vision system for classifying industrial parts with explainability.

---

Overview

SmartSort-CAM is a modular CV system built for industry-grade inspection workflows. It combines:

• Synthetic data generation using Blender
• ConvNeXt-based classification
• Grad-CAM explainability
• FastAPI inference server
• Docker containerization

---

Repository Structure

Refer to the README for full details on each folder’s responsibility.

---

Key Features

• High-fidelity synthetic rendering (HD lighting, metallic texture, randomized views)
• ConvNeXt-Large classifier trained on synthetic parts
• REST API interface for deployment
• Grad-CAM overlays for explainability
• Inference tracking + automation using Docker and Makefile

---

Getting Started

1. Clone the repository:

   git clone https://github.com/gbr-rl/SmartSort-CAM
   cd SmartSort-CAM
   

2. Build the container:

   make build
   

3. Run with GPU support:

   make run-gpu
   

4. Send image to API:

   python app/main.py path/to/image.png
   

---

Model Performance

The ConvNeXt model was trained on high-quality synthetic images and achieved:

• 100% accuracy on validation set
• 99.5% accuracy on shifted-domain images (blurred/low-light)
• Precision, Recall, F1: All = 1.0

---

Contact

Built by Rohiith Gettala, MSc Robotics, RWTH Aachen University.
Passionate about computer vision, synthetic data, and real-time inference systems.